The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires downloading and installing code from a repository on GitHub (github.com/CTCT-CT2/openclaw-security-watchdog) that is not associated with a known trusted organization or the skill's primary vendor. This creates a supply chain risk.
[DATA_EXFILTRATION]: When operating in 'Full Scan' mode, the tool sends system metadata and performs lookups against an external API (threat-intel.openclaw.io). This involves transmitting information about the host system's security posture to a third-party service.
[PRIVILEGE_ESCALATION]: The documentation instructs the user to use sudo to bypass permission restrictions on sensitive files like /var/log/auth.log and /etc/ssh/sshd_config. It also recommends running the main application with root privileges, which significantly increases the risk if the tool were to be compromised.
[COMMAND_EXECUTION]: The installation and usage instructions rely on the execution of shell commands, including npm install for dependencies and sudo for system-wide auditing. These commands have the capability to modify system configuration and environment settings.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from system logs, file permissions, and external threat intelligence APIs. 1. Ingestion points: Reads from /var/log/auth.log, system configuration files, and API responses from threat-intel.openclaw.io. 2. Boundary markers: No explicit sanitization or prompt delimiters are mentioned to prevent malicious log entries from influencing the agent's behavior. 3. Capability inventory: The skill has significant capabilities, including network requests, file reading/writing, and shell command execution. 4. Sanitization: There is no evidence of filtering or escaping external data before it is processed or reported to the user.

openclaw-security-watchdog